Hyperspectral Leaf Image-Based Cucumber Disease Recognition Using the Extended Collaborative Representation Model

1. Introduction

Plant diseases severely threaten the yield and quality of agricultural products. Rapid, accurate, and reliable disease detection and identification is vital to disease prevention and control for sustainable agriculture and food security [1]. Traditional methods rely on agronomists manually checking the plant disease symptoms or visible signs of a pathogen with the naked eye [2,3] or professional analysts performing physiological and biochemical analysis including molecular, serological, and deoxyribose nucleic acid [4,5]. Meanwhile, the visual assessment method requires plant to show visible symptoms, which is often used in the middle to late stage of infection [2]; besides, the diagnostic result is heavily influenced by the subjective consciousness and empirical knowledge of observers. As for the method of physiological and biochemical analysis, it is time-consuming and labor-intensive [6], and specific operating environment as well as high level of expertise and operating skills of the analyst are highly demanded to obtain reliable diagnosis results.

With the rapid development of computer vision and artificial intelligence, image processing techniques have shown great potential in automatic disease diagnosis, which can overcome some defects of the above methods and mitigate the problem of lack of expertise in the field of agriculture [7]. By now, numerous image processing-based diagnosis methods or systems have been developed by researchers and have achieved great success [1,8,9,10,11,12,13]. For instance, based on image processing techniques and artificial neural networks, Pawar et al. [1] proposed a real-time cucumber disease detection system that consisted of five sequential procedures, including image acquisition, preprocessing, feature extraction, creating database and classification, providing classification accuracy of 80.45% on cucumber downy mildew, powdery mildew, and healthy plants. Zhang et al. [9] segmented diseased blade images by the K-means clustering method, extracted the shape and color features from the lesions, and utilized the sparse representation classifier to achieve rapid identification of cucumber diseases. Based on leaf images, Sladojevic et al. [10] utilized deep convolutional neural networks to distinguish 13 different types of diseases out of healthy leaves and achieved precision between 91% and 98%. In reference [11], Ferentinos trained several convolutional neural network models using a large open database containing of 58 classes, and realized disease diagnosis using simple blade images from healthy and diseased plants. Jia et al. [12] segmented blade images by...